MORE than 100 humanoid robots are currently training to perform highly challenging precision tasks at X-Humanoid’s Embodied Intelligence Robot Data and Training Base in Beijing. The training relies on simulation data generated by on-site operators, helping machines acquire the skills needed for complex real-world applications.

Wearing virtual reality headsets or using remote operation devices, workers perform movements across a range of scenarios — including home, health and wellness, shopping, industry, and medicine — which are then converted into training data for the robots.

In a home setting, one robot carefully washes dishes, while in a health and wellness scenario, another gently tucks in a dummy, and a separate robot practices changing a baby’s diaper. In industrial training, a robotic arm learns to sort parts and twist screws, as part of preparation for tasks such as electric power inspections.

According to a staff member, this immersive training process goes beyond simple remote control; it serves as “data feeding.” The base is equipped with over 120 devices capable of generating 400 hours of data per day. All of this data is used to support the internal algorithm team as well as external partners, including robot manufacturers and artificial intelligence model developers.

The data collected at the base includes precise details that are difficult to replicate through simulation alone, such as force feedback, tactile information, and environmental interference. A staff member noted that this type of “physical intuition” can only be developed through multimodal data gathered by real robots.

Three months ago, the base’s data qualification rate stood at just 50%, according to Jiang Weilai, the facility’s head. At that time, challenges included overly bright lights causing overexposure in images, as well as robots accidentally making contact with props outside their designated areas.

Following months of personnel training, process refinement, issue tracing, and quality standard optimization, the data qualification rate has now stabilized at 95%, Jiang said.(SD-Agencies)